Using Adaptive Compilation to Produce High Performance Sparse Computations

Description

Matrix operations appear frequently in many areas of science and engineering. Often,
the solution of real life problems requires efficient computations on sparse matrices. The
performance of a code which operates on sparse matrices depends on many parameters,
such as the density and sparsity pattern of the input matrix, the data structures used, or
the processor's cache sizes, among others.

We are interested in exploring the adaptive compilation approach with the aim to
produce high performance implementations of codes working on sparse matrices. Using
an iterative compilation approach we will produce high performance implementations of
some operations which deal with sparse matrices. In addition, we will explore the
possibility to select which version to use at run time depending on the input problem.

Current Researchers

* Georgios I. Goumas, Computing Systems Laboratory, Athens, Greece
* José R. Herrero, Universitat Politècnica de Catalunya, Barcelona, Spain

This cluster welcomes other members interested in the above topic.